The host-defense peptide piscidin P1 reorganizes lipid domains in membranes and decreases activation energies in mechanosensitive ion channels

The host-defense peptide (HDP) piscidin 1 (P1), isolated from the mast cells of striped bass, has potent activities against bacteria, viruses, fungi, and cancer cells and can also modulate the activity of membrane receptors. Given its broad pharmacological potential, here we used several approaches to better understand its interactions with multicomponent bilayers representing models of bacterial (phosphatidylethanolamine (PE)/phosphatidylglycerol) and mammalian (phosphatidylcholine/cholesterol (PC/Chol)) membranes. Using solid-state NMR, we solved the structure of P1 bound to PC/Chol and compared it with that of P3, a less potent homolog. The comparison disclosed that although both peptides are interfacially bound and α-helical, they differ in bilayer orientations and depths of insertion, and these differences depend on bilayer composition. Although Chol is thought to make mammalian membranes less susceptible to HDP-mediated destabilization, we found that Chol does not affect the permeabilization effects of P1. X-ray diffraction experiments revealed that both piscidins produce a demixing effect in PC/Chol membranes by increasing the fraction of the Chol-depleted phase. Furthermore, P1 increased the temperature required for the lamellar–to–hexagonal phase transition in PE bilayers, suggesting that it imposes positive membrane curvature. Patch-clamp measurements on the inner Escherichia coli membrane showed that P1 and P3, at concentrations sufficient for antimicrobial activity, substantially decrease the activating tension for bacterial mechanosensitive channels. This indicated that piscidins can cause lipid redistribution and restructuring in the microenvironment near proteins. We conclude that the mechanism of piscidin's antimicrobial activity extends beyond simple membrane destabilization, helping to rationalize its broader spectrum of pharmacological effects.

tively or individually, demonstrate pH-resiliency despite being particularly rich in histidine (20% versus 2% on average in other AMPs), a residue known to be responsive to pH in the physiological range. In particular, the membrane activity of P1 is pHresilient (8). For these reasons P1 represents a promising template for molecular therapeutics, but future developments require that we examine in greater detail the physicochemical properties that make P1 an effective antimicrobial, anticancer, and antiviral peptide.
By analogy with other HDPs, the direct cytotoxicity and antimicrobial activities of P1 are thought to derive primarily from its ability to perturb the lipid bilayer structure of cell membranes. Indeed, there is no evidence that direct bacterial killing by P1 involves binding to a specific membrane receptor site. Similar arguments have been made for other HDPs (9) as well as tarantula neurotoxins (10) because synthetic enantiomers of these peptides exhibit identical antibacterial or ion channel modulation abilities, and thus they do not display the features associated with the molecular recognition of chiral protein receptors. Interestingly, P1 exhibits strong anti-inflammatory and anesthetic properties (11), indicating that it has immunomodulatory effects on host cells. Furthermore, both P1 and its less potent piscidin homolog, P3, exclusively activate formyl peptide receptors 1 and 2 on the surface of neutrophils and induce chemotaxis through these two G-protein-coupled receptors (12). Importantly, a common denominator for these multifaceted actions of P1 is the lipid bilayer of the various bacterial, viral, and mammalian cells that the peptide recognizes and targets.
The higher specificity of cationic HDPs toward bacterial versus mammalian cell membranes has been broadly attributed to their electrostatic interactions with anionic lipids, such as phosphoglycerol (PG) and cardiolipin (CL), that are abundant in bacterial membranes (13)(14)(15). Conversely, it has been observed that the presence of Chol in cell membranes typically inhibits the binding and lytic activity of several HDPs (16 -18). The common reasoning for this is that Chol rigidifies the bilayer and increases its hydrocarbon thickness, thereby preventing HDP insertion. However, P1 is highly active against many Chol-rich cells, such as human cancer cells (5) and HIV-1 (4). Apart from the presumed protective role of Chol to increase bilayer order and preserve the structural integrity of the membrane in the presence of HDPs, numerous studies suggest that Chol segregates within "lipid rafts," i.e. specialized types of membrane domains that are important for cellular function (19). Although direct detection of such domains in natural membranes is a challenge, indirect evidence of their existence and the role in protein function exists. For instance, recent investigations using super-resolution imaging showed that activation of TREK-1 channels depends on raft disruption via mechanical stress or anesthetic action in cellular membranes (20,21). Here, we propose that, similar to other membranesoluble anesthetic molecules and analgesic toxins (10), P1 exerts some of its actions, including its anesthetic effects (11), by causing topological changes and redistribution of liquid-ordered (L o )-liquid-disordered (L d ) phases in membranes (22) leading to mechanical stress redistribution and modulation of membrane protein function (23).
Despite their simpler composition compared with their mammalian counterparts, bacterial cell membranes also exhibit organization in domains (24). Notably, P1 and its close homolog, P3, were found to concentrate at septal regions (25), which are known for their role in specific cellular processes (e.g. cell division and sporulation) and subcellular localization of membrane proteins (e.g. enzymes involved in the synthesis of PE, CL, and PG lipids) (24). Importantly, several studies have demonstrated that septal regions are rich in nonlamellar-forming lipids such as PE and CL (Ref. 15 and references therein). Apart from sustaining membrane proteins, these structurallylabile regions must allow extreme topological bilayer transformations, thereby rendering them particularly susceptible to the disruptive effects of HDPs.
Given that the impact of sequence variability on the interactions of P1 and P3 with mimics of bacterial cell membranes was previously addressed (8,25,26), this article focuses on the more membrane-active member, P1, and P3 is used for comparative purposes. With the main goal of better understanding the modes of action of P1 in heterogeneous membranes, we investigated the interactions of this peptide with a few multicomponent membrane systems, including mixtures of PC with Chol, PE with PG, as well as the natural inner cell membrane of Escherichia coli. We present structural and functional experimental data that reveal how P1 exploits lipid domain formation for its multifaceted action in heterogeneous membranes, including changing lipid domain distributions and ion channel activities. This new knowledge helps us better understand the broad range of P1 biological activities.

Permeabilization assays of POPC/Chol vesicles in the presence of P1 and P3
Membrane permeabilization by membrane-active peptides is well-known to occur in a concentration-dependent fashion, with the threshold concentration for activity correlating with the reorientation and deeper insertion of the peptides in membranes (27)(28)(29). We prepared 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and 4:1 POPC/Chol large unilamellar vesicles (LUV) containing trapped calcein and measured the ability of P1 and P3 to release the fluorescent dye as a function of the peptide-to-lipid molar ratio (P/L). The assays were done at pH 7.4 when the histidine side chains are neutral (8). As shown in Fig. 1, P1 and P3 both permeabilize the two types of LUVs, and the threshold at which permeabilization occurs is characterized by a relatively high error bar compared with the other data points, presumably due to the stochastic aspects of the process that underlies leakage (30,31). P1 is significantly more effective than P3 in both lipid systems, and the addition of Chol does not affect the effectiveness of the peptides. Indeed, the P/L producing 50% permeabilization (EC 50 ) of the POPC LUVs is lower for P1 (1:166) than P3 (1:28) and comparable within the experimental error of 20% whether POPC contains Chol or not (1:130 for P1 and 1:23 for P3). In 3:1 POPC/POPG, the EC 50 values were P1/L ϭ 1:22 and P3/L ϭ 1:4 (8), whereas in 1:1 POPE/POPG, they were P1/L ϭ 1:10 and P3/L ϭ 1:3 (25). Hence, the peptides appear to be more membrane-active in bilayers lacking anionic lipids, possibly because the anionic component prevents the peptides from inserting as deeply as in the zwitterionic bilayers, an effect observed for other HDPs (32).

High-resolution structures of P1 and P3 in POPC/Chol
Although the antimicrobial peptide data bank contains more than 3,000 peptides, only 13% have known three-dimensional (3D) structures (33). This is partly explained by the difficulty associated with solving the structures of amphipathic peptides that do not form crystals suitable for structural determination by X-ray diffraction (34,35). We previously took advantage of oriented sample solid-state NMR to obtain both the structures and orientations of P1 and P3 under native-like conditions, i.e. in the presence of hydrated, fluid phospholipid bilayers (26). More precisely, these structures were solved in the presence of 3:1 1,2-dimyristoyl-sn-glycero-3-phosphatidylcholine/1,2dimyristoyl-sn-glycero-3-phosphatidylglycerol (DMPC/DMPG) and 1:1 POPE/POPG bilayers as mimics for Gram-positive and -negative bacterial cell membranes (26). Rigorous structural determination of peptide structures in the presence of fluid bilayers requires incorporating the effect of peptide dynamics on the NMR restraints (36 -39). As shown for the piscidin structures solved in DMPC/DMPG and POPE/POPG, 15 N anisotropic chemical shifts (CSAs) and 15 N-1 H dipolar couplings (DCs) extracted from two-dimensional spectra of 15 Nlabeled piscidins provided accurate structural restraints that are highly sensitive to not only secondary structure but also the orientation of the peptides with respect to the bilayer normal. In particular, the 15 N-1 H DCs exhibit a strong dependence on the tilts of the helical axis () and average azimuthal () angles adopted by the piscidin peptides bound to bilayers, and thus they can be used to reveal their kinked structures. Furthermore, we demonstrated that all structural restraints are accurate despite peptide dynamics. Here, we solved the structures of P1 and P3 in a binary 4:1 POPC/Chol mixture, as a way to mimic the zwitterionic nature and cholesterol content found in the outer leaflet of mammalian cell membranes (40).
Following the same approach as used previously (26), we obtained 15 N-1 H DCs and 15 N CSAs for P1 and P3 bound to 4:1 POPC/Chol by performing the two-dimensional heteronuclear correlation (HETCOR) experiments (41) on multiple 15 N-labeled peptides bound to oriented bilayers. The spectra collected for several multiple-labeled samples of P1 were superimposed to generate Fig. S1. Fig. 2A and Fig. S2A show the lowest energy structures that were calculated upon refinement of the NMR restraints collected for P1 and P3 in 4:1 PC/Chol, at P/L ϭ 1:40. Table S4 summarizes the statistics for the calculated structures. The RMSDs for heavy atoms, which are 1.16 and 1.12 Å for P1 and P3, respectively, are based on considering the top 10 structures and focusing on the residues that exhibit ␣-helicity without fraying, i.e. residues 3-20. These RMSD values demonstrate the excellent agreement between the different restraints used for the structural determination.
The structures of P1 and P3 in PC/Chol underscore several important features. First, similarly to the structures determined in PC/PG and PE/PG, those obtained in PC/Chol exhibit highly ␣-helical and kinked structures, indicating that secondary structure is not significantly affected by membrane composition (26). Kinking in the middle of the peptides is mostly due to the rotation of the helix (characterized by the angle) being different on each side of the conserved glycine at position 13. Such structural imperfection enables the peptides to optimize their hydrophobic moment in the presence of the bilayer (26). As shown in Fig. 2C and Fig. S2C, the ␣-helix of P1 is more rotated at its N-than C-terminal end, as reflected by larger N than C values, respectively (Table S5). Furthermore, P1 adopts a significantly larger N value than P3. Hence, is an orienta-  Table S1 and compared with previously obtained values in 1:1 POPE/POPG. The % calcein leakage (mean Ϯ S.D.) is displayed as a function of L/P1 and L/P3 for at least six measurements (n ϭ 6). . The NMR samples were prepared at pH 7.4 using a P/L of 1:40. The structure, which represents the lowest-energy member of the ensemble of structural conformers, is displayed for a peptide partitioned in the upper leaflet of the bilayer. Hence, the basic (stick representation) and hydrophobic side chains point upward and downward, respectively. Gray lines represent the average position of the C2 atoms of the lipids, based on prior molecular dynamic determinations (42), and yellow circles are shown to represent the position of the peptide center of mass. The RMSD between the top 10 structures is 1.14 Å in the ␣-helical region that experiences no fraying (residues 3-20) (see Table S4). B, structure of P1 as solved previously in 1:1 POPE/POPG (PDB code 2MCV) (26). C, helical wheel diagram of P1 in 4:1 POPC/ Chol. The kink at a central glycine, Gly-13, allows the two halves of the helix to rotate independently around the helical axis, as indicated by different N and C values, for the N-and C-terminal regions, respectively (26). Red arrows and values represent the rotation angles for P1 in POPE/POPG for comparison, and the orange arrow represents the direction of the hydrophobic moment (H).

Activity of piscidins at heterogeneous membranes
tional characteristic that is particularly pertinent to capturing not only the kink at Gly-13 but also the different bilayer arrangements of P1 and P3.
Second, a recurring feature of the bilayer-bound piscidin structures is that, due to subtle changes in the rotation of the helix in the membrane, the side chains are arranged to minimize the footprint of the peptide on the bilayer plane (Table S6 and Fig. S10). This is because most of the bulky hydrophobic residues are grouped together and point straight into the bilayer, although the large polar residues point toward the water phase, leaving the small Gly and Val residues to occupy the sides of the helix (Fig. 2C). This distribution gives the peptide the shape of a sharp wedge that may be used to facilitate immersion into the bilayer.
Third, comparing the structures of P1 and P3 in POPC/Chol (4:1) at P/L ϭ 1:40 ( Fig. 2A and Fig. S2A) with those in POPE/ POPG (1:1) at P/L ϭ 1:20 ( Fig. 2B and Fig. S2B) reveals that the peptides adopt a more pronounced tilt in the zwitterionic lipids. For P1, it is the C-terminal region that is more tilted in PC/Chol (80 Ϯ 1°) than PE/PG (86 Ϯ 1°), as shown in Fig. 2 A and B. In contrast, it is the N-terminal end of P3 that is more inclined in PC/Chol (85 Ϯ 1°) than PE/PG (92 Ϯ 1°) (Fig. S2, A and B). We previously published molecular dynamics data showing that P1 and P3 insert more deeply in the 4:1 POPC/ Chol than 1:1 POPE/POPG bilayers (42). Indeed, the depth of insertion for the center of mass of P1 was found to sit 1.0 Ϯ 0.2 and 0.6 Ϯ 0.3 Å below the C2 of the lipid acyl chains in 4:1 POPC/Chol and 1:1 POPE/POPG bilayers, respectively. We used those as guiding values for positioning our structures in Fig. 2. Because the peptides are more lytic in the zwitterionic bilayers, the subtle adjustments in helix rotation and the increased tilting and depth of insertion revealed by our highresolution structural studies hint that these properties are important for increased membrane activity, as discussed below.

X-ray diffraction studies of POPC/Chol in the presence of P1 and P3
We investigated lamellar samples containing P1 or P3 in the binary mixture of POPC and Chol at various P/L values and POPC/Chol molar ratios by X-ray diffraction. As reported previously (43), PC/Chol binary mixtures at Chol molar fractions of 0.5 or below do not show phase separation. Our data confirm that a mixture of 2:1 POPC/Chol, prepared as oriented multilayers, yields one set of equidistant diffraction peaks corresponding to a homogeneous phase at a single repeat spacing (Fig. 3A). Adding P1 or P3 to the POPC/Chol binary mixture causes the appearance of an additional set of Bragg peaks. Partitioning of amphipathic peptides at the bilayer-water interface is expected to thin the bilayer (Fig. S3) due to an area expansion at constant hydrocarbon density, a feature commonly observed in the presence of many membrane-active peptides (44 -46). However, Chol clustering with lipids causes lipid-chain ordering and stretching, resulting in a thicker bilayer (47,48). The mismatch in the thicknesses of the two coexisting phases (domains) in a bilayer together with stacking of like-domains across the multilayers give rise to separate sets of Bragg peaks, explaining the two distinct repeat spacings. The two phases, which differ in bilayer thicknesses by more than 2 Å (Fig. 3A, inset), can be described as an L o phase, rich in Chol ("C" phase), and an L d phase depleted of Chol and enriched with peptide ("L" phase). The C phase displays a slightly thicker bilayer in the presence of peptide, compared with neat (pure) POPC/Chol (Fig. 3B). This can be explained by an increased density of Chol in the C region, as Chol is pushed away by the peptide. In contrast, the electron density profiles of the peptide-perturbed L phase ( Fig. 3C) relative to the pure POPC reveal that the bilayer suffers massive perturbations in the presence of piscidins, with P1 being more disruptive than P3, based on the extent of smearing detected in the profiles. The disorder is so significant that

Activity of piscidins at heterogeneous membranes
the segregation of polar-nonpolar regions of the bilayer is almost lost. This is accompanied by a significant bilayer thinning compared with a neat POPC bilayer (Fig. S3).
To further investigate the appearance of two phases in the presence of piscidin, we performed experiments at other P/L and POPC/Chol ratios (Fig. S4, A and B). We noted that the phase separation persists even at a lower Chol fraction (e.g. 4:1 POPC/Chol) if enough peptide is present (P/L Ͼ1:25), but it is not observed at lower peptide fractions, including the P/L of 1:40 used for the solid-state NMR structural studies (Fig. S4B). This suggests that the separation occurs when the P/L and POPC/Chol reach specific threshold concentrations. Interestingly, at P/L ϭ 1:25, P1 is near the threshold concentration for ϳ100% dye leakage from POPC and POPC/Chol liposomes ( Fig. 1) but only near the leakage midpoint for the POPC/POPG liposomes (8). Thus, phase separation and maximum bilayer disruption appear to be correlated. Overall, these data indicate that both P1 and P3 trigger phase separation in POPC/Chol mixtures with preference for occupying the Chol-depleted L-phase of POPC, thus causing an effective increase in local peptide density. This, in turn, can exacerbate local bilayer deformations and peptide-induced permeabilization as indicated by our neutron diffraction (Fig. 3C) and dye-leakage ( Fig.  1) data, respectively.

Fluorescence microscopy of P1 and P3 in raft-forming mixtures
To confirm the possibility that piscidin preferentially partitions in the L d phase, we investigated the partitioning preferences of P1 and P3 in giant unilamellar vesicles (GUVs) prepared from raft-forming lipid/Chol mixtures. The fluorescent lipid FastDio was shown to preferentially partition in the L d phase (49). As shown in Fig. 4, the location of the TAMRAlabeled peptides (red) coincided with the FastDio-labeled lipid (green), confirming that both P1 and P3 colocalize with the L d phase.

X-ray diffraction studies of POPE/POPG mixtures in the presence of P1 and P3
PE and PG lipids are major components of the bacterial membranes, with PE constituting roughly 80% of E. coli phospholipids (50). Within the temperature range of 0 to 100°C, POPE shows two main transitions: a gel-to-fluid (L ␤ to L ␣ ) transition at 25°C and a fluid lamellar (L ␣ ) to inverted hexagonal (H II ) phase at around 75°C (Fig. S5A), whereas POPG is in a fluid lamellar (L ␣ ) phase. Our diffraction data from lamellar samples made of POPE and POPG show that at temperatures below 25°C, the gel phase of POPE separates from the fluid phase of POPG (Fig. 5A). This behavior changes dramatically in the presence of P1, as the multiple sets of peaks merge into one at temperatures as low as 15°C (Fig. 5B). The gel phase "melts" into a unified fluid phase at temperatures well-below the melting transition for pure POPE. A similar trend is found for P3 (Fig. S6); however, the phase mixing happens at slightly higher temperatures, indicating that P1 is more efficient in altering the phase state behavior of POPE. To uncover the connection between phase behavior and physical location of the peptide in these type of bilayers, we performed differential scanning calorimetry (DSC) measurements coupled with neutron diffraction in PE lipids, as described below.

Differential scanning calorimetry of POPE/POPG and diPoPE in the presence of P1 and P3
Phosphatidylethanolamine (PE) lipids are characterized by cone-shaped molecular geometry, due to the small PE headgroup area compared with the acyl tails. In addition, the PE headgroup can form extensive intermolecular hydrogen bonds with other PE molecules (51). Hence, this lipid possesses a large negative spontaneous curvature (52,53). These properties make PE lipids prone to packing into a tight gel phase at low temperatures and forming nonlamellar structures at higher temperatures. A gel-to-fluid phase transition is found at T m ϭ 25°C for POPE, and at 20°C for 3:1 POPE/POPG (Fig. S5A).
Addition of P1 appears to cause "tailing" of this transition toward the low temperature side (Fig. 6A). A similar trend is found for P3 (Fig. S5B). Other HDPs in similar lipid mixtures were found to produce a splitting into two close transitions, presumably because the cationic peptides segregate with the anionic lipids (54,55). Although the lipid segregation is not obvious from our DSC scans for P1 (P3), we do notice a broadening of this main transition in the presence of peptides, when compared with the pure POPE/POPG lipid ( Fig. 6A and Fig.  S5B). This may explain the accelerated melting of POPE in the presence of piscidin, similar to that caused by increasing the sample temperature.
Because POPE shows an L ␣ to H II transition at T H ϭ 75°C, which is far from physiological temperatures, dipalmitoyl-PE (diPoPE, T H ϭ 43°C) can be used for a more amenable detection of this phase transition (56). When we added small amounts of P1 to diPoPE (P/L ϭ 1:350), we detected a strong L ␣ to H II transition that occurred 5°C higher compared with pure diPoPE (Fig. 6B). The difference suggests that the peptide

Activity of piscidins at heterogeneous membranes
imposes positive curvature strain, opposing the intrinsic negative curvature of the lipid, thus delaying the transition to the hexagonal phase. The types of interactions that dominate the association of lipids in the bilayer are as follows: water repulsion from the hydrocarbon region; van der Waals between the hydrocarbon chains; tight solvation of the PE headgroups; and hydrogen bonding between headgroups. Notably, the enthalpy of the L ␣ to H II transition is larger in the presence of P1 compared with the pure lipid (Fig. 6B), indicating that P1 affects the forces acting between the PE lipids. The following question then arises. How would P1 distribute in the bilayer to create such an effect? To answer this question, we employed neutron diffraction and peptide deuterium labeling, as described below.

Neutron diffraction profiles of diPoPE bilayers with deuterated P1
We incorporated a deuterated form of P1 (d 33 (8) in lamellar samples of diPoPE containing a small amount (5% molar) of POPG or d 31 -POPG. The incorporation of d 31 -POPG was needed for a quantitative analysis of the water content by neutron diffraction (Fig. S7) (57). A pair of samples containing P1, in either unlabeled or deuterated (d 33 -P1) form, were prepared at the same time. The positions and distributions of the deuterated components in the bilayer were calculated using deuterium contrast (58). This included determining the water profile, via H 2 O/ 2 H 2 O exchange. Fig. 7 shows the resulting deuterium profiles of P1 label (d 33 ) and water ( 2 H 2 O) relative to the overall profile of the diPoPE bilayer containing P1 in a nondeuterated form. Only one broad deuterium peak, positioned superficially, in the PE headgroup region, can be distinguished. Although two sites on P1 were deuterated, near the N and C termini, the two sites cannot be parsed out in the profile, indicating that the peptide is oriented roughly parallel to the membrane surface. This is to be contrasted with our previous results for P1 in POPC/POPG where a pronounced penetration and tilt in the bilayer could be identified from the distinct positions of the same two deuterated regions (8).
The superficial location of P1 in diPoPE indicates that the interaction is concentrated in the lipid headgroup area. Water colocalizes with the lipid headgroups and the peptide at the water-bilayer interface (Fig. 7). All our studies indicate that P1 has a higher propensity to tilt, insert into, and permeabilize PCversus PE-containing bilayers. Conceivably, contributing factors include the larger area per headgroup for PC versus PE (by about 10 Å 2 , at full hydration) (59) and the higher headgroup hydration, both of which could facilitate the integration of the amphipathic peptide in the bilayer. Indeed, using deuterium for calibration, we determine here that 8.1 waters associate with

Activity of piscidins at heterogeneous membranes
each PE headgroup in diPoPE, compared with 9.4 waters found previously for DOPC (60), both determined at 93% relative humidity. PE has a primary amine in its headgroup, making it capable of forming both intra-and inter-molecular hydrogen bonds. This results in a more densely-packed lipid-water interface compared with a PC membrane, and less room for water and ions to bind. Because P1 resides on the bilayer surface in diPoPE (Fig. 7), it participates in the hydrogen-bonded network with PE headgroups and water. In the H II phase, it was shown that POPE lipid headgroups wrap around water-filled cylindrical channels that are roughly 30 Å in diameter (61). Such channels could accommodate at least one peptide with the helical axis oriented along the cylindrical axis (Fig. 6B, inset). To allow a change in membrane surface topology, as would be the case in a L ␣ -to-H II transition, the peptide would need to re-orient its helical axis along the cylindrical axis. Such a peptide re-arrangement could be energetically costly, as suggested by the higher enthalpy for this transition, compared with the pure lipid (Fig. 6B). Taken together, our data show that P1 intercalates between the relatively small PE headgroups disrupting the lipid packing (downward shift in T m ) and opposing the natural tendency of the diPoPE to curve (upward shift in T H ).

Patch-clamp experiments using E. coli spheroplasts in the presence of P1 and P3
The above-observed shifts in the lipid phase behavior and domain distributions in membranes upon interaction with P1 create significant changes in the lateral pressure profiles. Such disruptions could affect ion channel behavior in both bacterial and mammalian cell membranes. To explore the possibility of such an effect with P1 and P3, we investigated the behavior of mechanosensitive channels in E. coli spheroplasts, treated with both peptides (Fig. 8 and Fig. S8). The integral membrane pro-teins that are intrinsically designed to sense lateral pressure/ tension are called mechanosensitive (MS) channels. MscS and MscL channels, which represent the two most understood tension-gated bacterial osmo-regulatory valves (62), are well-characterized, and thus convenient to detect possible perturbations of the lateral pressure profile in the inner bacterial membrane, where most HDPs deploy their membrane activity. Fig. 8A shows the typical response of a native MS channel population to a linear ramp of pipette pressure (suction). The control curve illustrates two waves of electrical activity: the first wave reflects activation of the low-threshold MscS channels, which saturates, and the second wave represents the population of high-threshold MscL channels. If the shape and curvature of the patch stay constant in the range of activating pressures, the midpoint pressures (p 0.5 ) for the MscS and MscL populations directly reflect activating tensions (63). As shown in Fig. 8, the ratio of p 0.5 (MscS) to p 0.5 (MscL) is close to 0.6 in the absence of P1. Perfusion of P1 (1.0 mol/liter) in the chamber bathing the cytoplasmic side of the patch followed by a 15-min equilibration period reproducibly led to a reduction of activation midpoints for both channels. Fig. 8B shows values of pressure midpoints normalized to the p 0.5 (MscL) recorded in response to the first ramp (pull) in the absence of peptide. The second and third ramps were applied to make sure that the patch was stable and that the midpoint did not change substantially with time. After the third ramp, P1, was applied to the bath and after a 15-min equilibration, three more sequential ramps were applied. Some patches mechanically broke under the fifth and the sixth ramp application, and for this reason the number of points on the graph decreased with the number of pulls.
The major information gained from these experiments is that the average mid-point values between the third and fourth pulls decrease in the presence of P1 (Fig. 8B). Indeed, the average relative midpoint position shifted from 0.98 Ϯ 0.02 (in the absence of P1) to 0.78 Ϯ 0.09 (in the presence of P1) for MscL and from 0.59 Ϯ 0.01 to 0.50 Ϯ 0.04 for MscS (n ϭ 7). Because the tension midpoint and the in-plane expansion of the channel complex directly reflect the free energy of the opening transition (63), we conclude that for each of these channels the effective transition energy decreased by ϳ20% in the presence of P1, signifying a substantial change in the way the forces in the lipid bilayer are conveyed to the channels. Fig. S8 illustrates similar experiments performed with P3. Importantly, in all cases we see similar two-wave activation curves and clear unitary MscS currents at the foot of each activation curve signifying that in these curves mechano-activated channel currents are not intermixed with conductances of piscidin-produced pores. The latter appear in E. coli patches at substantially higher voltages (Fig.  S9). Both peptides showed comparable effects of midpoint reduction at 1.0 mol/liter. Lower peptide concentrations (0.1-0.5 mol/liter) produced less reproducible shifts, whereas all tested patches ruptured under mechanical stimulation in the presence of 2 mol/liter of either peptide, as both exert strong membrane destabilization. These experiments demonstrate the ability of both P1 and P3 to sensitize mechanosensitive channels in bacterial cell membranes, effectively decreasing the energy input of external tension required for the opening tran-

Activity of piscidins at heterogeneous membranes
sition (63). The observed effects occur at a concentration relevant to the antimicrobial activity of the peptides.

Discussion
Natural membranes are constituted from a variety of lipid species, and their compartmentalization in domains have important biological functions (19, 24). Membrane heterogeneity has received little attention when discussing the mechanisms of action of HDPs. However, it has been observed that anionic lipid clustering caused by cationic HDPs and cell-penetrating peptides can contribute to their mechanism of action (64). P1 differs from other well-studied HDPs in that it exhibits a high adaptability to pH, salinity, and various lipid environments (1,(4)(5)(6). This versatility derives partly from P1's capacity to regulate charge across its four histidine residues, thus controlling its hydrophobicity (8) and, as we propose here, its ability to re-organize membrane microenvironments in either bacterial or mammalian cells, resulting in multifaceted modes of membrane disruption.
As part of this study, we solved the high-resolution structures of P1 and P3 bound to fluid bilayers of 4:1 PC/Chol. This information is essential to testing new hypotheses about the modes of action of HDPs and designing novel therapeutics. Although P1 was found to be fully ␣-helical in SDS micelles (65) and only 45% structured in dodecyl phosphocholine micelles (66), our studies in native-like bilayers confirmed the trend previously obtained in 1:1 PE/PG and 3:1 PC/PG (26); the peptides are highly ␣-helical and are generally straight, but they are frayed at their extremities and have a kink described by a 25°rotational change between their N-and C-terminal domains. In a recent investigation (8), we showed that the contrasted histidine content of P1 and P3 correlated with their different directionality of membrane insertion, tilts, insertion depths, and membrane permeabilization effects in PC/PG bilayers. Overall, our multiple studies of two homologous peptides in different lipid environments highlight that stronger membranolytic effects are associated with increased tilting and insertion depth and optimization of the helix rotation in the membrane. Importantly, these three properties (tilt, depth of insertion, and helix orientation) vary as a function of the amino acid composition of the peptides and the composition of the membranes.
Although the formation of secondary structure is a major energetic driving force for the binding of amphipathic peptides to membranes (67), flexing at the central Gly-13 further improves amphipathicity. This maximized amphipathicity together with the high ability of P1 for charge regulation associated with its multiple histidines (8) allow the peptide to strongly anchor itself at the hydrophobic-hydrophilic interface in various lipid systems, independently of the presence of anionic lipids. It is, however, interesting to note that P1 permeabilizes membranes with equal efficacy in zwitterionic membranes whether Chol is present or not (Fig. 1), despite the presumed protective role of Chol against lysis. How can this be explained?
Our investigations in lipid membranes of POPC and Chol, major components of the outer leaflet of mammalian cell membranes, clearly show that P1 causes Chol to separate from a POPC/Chol binary mixture by recruiting phospholipids into a fluid phase (Fig. 3) or partitions exclusively into the disordered phase of a raft-forming mixture (Fig. 4). These reorganizations of the membrane can effectively boost the action of P1 because a higher concentration of peptide and larger deformations can occur in the Chol-depleted domains. As a result, the insertion and tilting of the peptide needed to elicit membrane disruption can occur at lower P/L than in homogeneous bilayers. Similar

Activity of piscidins at heterogeneous membranes
behaviors were observed for melittin (68) as well as for pardaxin (69) and scorpion HDPs (40) in binary POPC/Chol mixtures. L d domains in phase-separated (raft-containing) mixtures were shown to be targets for a diverse set of other antimicrobial peptides (70,71) and also to harbor both fiber and pore formation by the islet amyloid polypeptide (72). Thus, through their ability to induce phase separation even in simple binary POPC/Chol mixtures, piscidins can also create vulnerable sites of accumulation for other toxic peptides. The clear preference of P1/P3 to induce or partition into the L d phase indicates that the two piscidins prefer the phase where the hydrocarbon chains are more exposed. As we have shown previously, such exposure is reduced in the presence of Chol (47). Notably, the NMR structures reveal that piscidins adopt the shape of a sharp wedge in the bilayer environment, with side chains distributed in a way that minimizes the footprint of the peptide on the bilayer plane (Fig. S10). This allows the peptides to easily anchor themselves between the lipid headgroups at the hydrophilic-hydrophobic interface and more strongly so in the L d regions.
In the POPE/POPG mixtures used to characterize bacterial membranes, P1 and P3 have the effect of inhibiting the formation of the gel phase of POPE, thus altering the gel/fluid phase transition in a manner comparable with a significant increase in sample temperature. This effect is likely to interfere with the role of PE lipids as key regulators of bacterial membrane fluidity (73), especially in E. coli membranes that contain up to 80% PE lipids. Furthermore, our DSC results in diPoPE show that P1 also affects the bilayer morphology by imposing positive curvature strain. A similar effect was observed for the MSI-78 peptide (74) and LL-37 (75). Overall, the observed actions of P1 on PE bilayers, which includes loosening of the lipid packing (Fig. 5) and opposing the PE's intrinsic negative curvature (Fig. 6B), are likely to result in significant changes in the lateral pressure patterns in real bacterial membranes.
We show that in E. coli spheroplasts the action of P1 on the inner leaflet of the bacterial cytoplasmic membrane leads to the reduction of the midpoint pressures for the activation of both MscS and MscL channels, similar to the action of the ion channel toxin GsMtx4 (76). The lateral pressure profile is difficult to determine experimentally, and extensive molecular dynamics simulation would be needed to describe the local proteinbilayer interactions. Based on the data collected here, we propose two possible explanations for the observed effects. On the one hand, when P1 enters the annular layer of lipids around a mechanosensitive channel, it creates a substantial distortion of this layer, thus re-directing the external tension force that reaches the peripheral segments of the channel through protein-lipid interactions. This effect would likely occur due to the bilayer thinning and changes in the membrane intrinsic curvature causing the channels to perceive compressive forces normal to the plane of the membrane (77), thus leading to their activation at lower mechanical thresholds. Interestingly, addition of conical shape lipids such as lyso-PC to bilayers was found to dramatically lower the activation energies of the eukaryotic mechanosensitive channel (TREK-1 and TRAAK) (78) and drive the prokaryotic MscL into an open conformation (77). By analogy, P1 imposes positive curvature on PE bilayers, resulting in similar effects on the activation of McsL and MscS in the PE-rich E. coli membranes. On the other hand, if the peptide becomes a part of the channel-lipid boundary through direct interaction, it may increase the perimeter of this annular zone, thus increasing the total force acting on the channel (force is tension multiplied by perimeter). Effectively, both peptides decrease the energy of the closed-to-open transition by ϳ25% for MscS (i.e. from 24 to 18 kT (79)), and by 33% (from 58 to 38 kT) for MscL (80). The two proposed explanations are similar in the sense that they imply modification of the proteinlipid boundary (direct or indirect), thus re-directing forces acting from the bilayer to the protein. The peptide sub-lethal concentrations tested here lie just below the MIC ranges for E. coli (2-10 mol/liter for P1 and 10 -20 mol/liter for P3). Given the relatively large size of the permeation pores of MscL and MscS, loss of osmolytes through open ion channels can already occur at sub-lethal concentrations and before any peptide forms leakage-competent defects, resulting in bacterial growth inhibition. This may partly explain our previous findings that significant leakage of a small sugar analog molecule occurs through live E. coli membranes even at minimal P1 concentrations, well below 1 mol/liter (25).
Clearly, membrane heterogeneity plays an important role in the overall action of HDPs by creating the ground for preferential localization of HDPs in functionally important membrane regions (e.g. regions of high curvature stress and line boundaries) and opportunities for entry and interference with normal cellular processes. Through the examples of the piscidins P1 and P3, and our results in model lipid membranes, we provide evidence that HDPs are able to exploit the heterogeneity of membranes, or otherwise modify the membrane microenvironment to an extent that impairs function of membrane proteins, thus exhibiting multifaceted modes of action against invading cells. Overall, we find that the effects of the piscidins in either cholesterol-rich mammalian or PE-rich bacterial cell membranes feature, as a common ground, the strong promotion of the disordered phase. The resulting changes in lateral pressure profiles (23,81) can be significant enough to affect the conformations and functional behaviors of transmembrane proteins, including ion channels, and therefore they could offer a possible explanation for the observed anesthetic effect attributed to P1 (11). Furthermore, we show that shifts in the L o /L d phase distribution under the action of P1 and P3 can, in turn, influence their permeabilization properties, even at sub-lethal concentrations. This is likely an important but often overlooked mechanism of action for membrane-active HDPs.
Piscidins are especially interesting examples of HDP that show great adaptability, and therefore they may be a good starting model for the design of multipotent peptide treatments. Notably, P1 is very potent against a few lines of human cancer cells (5), many of which are known to contain increased levels of Chol, suggesting a potential use of P1 as a raft-modulating peptide agent for anti-cancer drug development (82)(83)(84).

Materials
Carboxyamidated P1 (FFHHIFRGIVHVGKTIHRLVTG-NH 2, M r 2,571) and P3 FIHHIFRGIVHAGRSIGRFLTG-NH 2 , Activity of piscidins at heterogeneous membranes M r 2,492) were used in all experiments. Unless otherwise indicated, they were obtained from Biomatik USA, LLC (Wilmington, DE) at a purity higher than 98%. Received as hydrochloride salts, the peptides were dialyzed against pure water, and the final concentrations were determined by amino acid analysis. The peptides used in the dye leakage assays, the 15 N-labeled peptides used in the NMR experiments, and the 2 H-labeled form of P1 (d 33 -P1 ϭ I5d 10 F6d 5 L19d 10 V20d 8 ) utilized in the neutron diffraction experiments were chemically synthesized at the University of Texas Southwestern Medical Center and purified as reported previously (26). After lyophilization, these peptides were dissolved in dilute HCl and dialyzed to substitute chloride for trifluoroacetate ions, leading to 98% pure peptides. Following reconstitution of the peptides in nanopure water, their molar concentrations were determined by amino acid analysis performed at the Protein Chemistry Center at Texas A&M. Chol (Ͼ 99% pure) was purchased from Sigma. Phospholipids were obtained from Avanti Polar Lipids (Alabaster, AL). These include POPC, POPG, d 31 -POPG, POPE, diPoPE, DOPC, 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC), 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC), and 1,2dioleoyl-sn-glycero-3-phospho-L-serine (DPPS).

Permeabilization assays
Calcein-loaded LUVs were prepared in the presence of P1 and P3, as described previously (25). LUVs contained 4 mol/ liter (total lipid) of 4:1 (mol/mol) POPC/Chol. The assays were performed in 96-well plates by pipetting 180 l of the LUV suspension and adding 20 l of the peptide solution. The final lipid concentration in each well was held constant at 10 mol/ liter, although the peptide concentration was varied to cover a range of P/L ratios between 2 and 256. Fluorescence was measured using a Varian (Walnut Creek, CA) Cary Eclipse spectrofluorometer. For the positive control, 20 l of 1% Triton X-100 was used in place of the peptides.
The HETCOR data were collected for P1 and P3 samples oriented with the bilayer normal is parallel to the static magnetic field, B 0 , yielding oriented 15 N CSAs and their associated

Neutron and X-ray diffraction
Lipids (2 mg) were dissolved in chloroform and mixed with peptides in trifluoroethanol (TFE) (Acros Organics) to the desired P/L. After evaporating the organic solvents under a flow of nitrogen, the samples were dried under vacuum for 1 h, thoroughly hydrated with nanopure water in a shaker at 35°C for 1 h, and then spread on thin glass coverslips. The bulk water was allowed to evaporate slowly overnight at room temperature. Before the diffraction experiments, the samples were annealed at 98% relative humidity and 30°C for at least 12 h. For additional controls, POPC/Chol mixtures without peptide were prepared in H 2 O as above, and water-solubilized P1 (P3) was subsequently added to the preformed lipid vesicles at the desired P/L. Samples containing diPoPE lipid were prepared directly from organic solvent because of their poor solubility in water, particularly at high concentrations. Deuterium-containing and natural abundance samples of P1 were prepared in parallel. Lamellar neutron diffraction sets, probing the direction orthogonal to the bilayer plane, were acquired with the instrument MAGIk at the National Institute of Standards and Technology Center for Neutron Research, Gaithersburg, MD. The data were processed and analyzed as described before (46,47). Tables with structure factors can be found in (Tables S2 and S3). Repeat spacings and their uncertainties were determined by a linear fit of the Bragg peak position versus diffraction order.
X-ray diffraction measurements were performed on a 3-kW Rigaku Smartlab diffractometer located at the Institute for Bioscience and Biotechnology Research (IBBR), Rockville, MD. Phases of the structure factors were determined by the swelling method (93). Structure factors were calculated from the integrated Bragg intensities after subtracting background and applying Lorentz, polarization, beam footprint, and absorption corrections. Electron density profiles were computed on an arbitrary scale, using direct Fourier reconstruction (94).

Fluorescence microscopy
GUVs were prepared at 84 mM in buffer (20 mmol/liter Tris, 50 mmol/liter NaCl, 127 mmol/liter sucrose, pH 7.4) using a previously described protocol (95,96). To yield liquid ordered and disordered domains, two lipid compositions were used: 17.  (96). The fluorescent lipid FastDio, which preferentially partitions in the L d phase, was added at 0.1 mol % (49). Following the formation of GUVs, 370 nmol/liter TAMRA-P1 or 540 nmol/liter TAMRA-P3 was added. Imaging was performed at 23°C on a Nikon Eclipse Ti microscope (Nikon Instruments, Melville, NY). Filters were used to avoid artifacts. Data were processed in ImageJ.

Differential scanning calorimetry
Samples were prepared as above by co-dissolving the peptide and lipid in TFE/chloroform. The organic solvent was removed under a stream of nitrogen gas, and placed under vacuum for 2 h. The dry lipid/peptide mixtures were resuspended in ultrapure water and allowed to hydrate overnight with continuous shaking. Alternatively, PIPES buffer was used (10 mmol/liter PIPES, 50 mmol/liter NaCl, phosphate, 0.5 mmol/liter EDTA, pH 7.4) for preparations of diPoPE samples, resulting in noisier data (Fig.  S5D). The samples were measured at a lipid concentration of 2.5 mg/ml. DSC measurements were made on VP-DSC microcalorimeter (MicroCal Inc., Northampton, MA). Six scans were made at a scan rate of 30°C/h. There was a 15-min equilibrating period prior to starting the experiment and a delay of 5 min between sequential scans to allow for thermal equilibration. DSC curves were analyzed by Origin, version 7.0 (OriginLab Corp.).

Patch-clamp measurements on giant E. coli spheroplasts
WT E. coli strain Frag-1, which natively expresses the mechanosensitive channels MscS and MscL as two dominant and readily observable species, was used in the patch-clamp experiments (97,98). Giant spheroplasts were prepared from Frag-1 cells using the standard steps of filamentous growth in the presence of cephalexin followed by cell wall digestion with lysozyme in the presence of EDTA, as described previously (79). Patch pipettes were pulled from borosilicate glass capillaries (Drummond Scientific no. 2-000-100) to the inner diameter of ϳ1.5 m and used without fire polishing or coating. All measurements were done in inside-out excised patches. Stimulating pressure protocols (linear suction ramps) were delivered from a pressure-clamp apparatus (ALA Instruments, Farmingdale, NY) and programmed in the PClamp-10 software (Molecular Devices, San Jose, CA). The standard spheroplast recording buffer contained (in mmol/liter) 200 KCl, 10 CaCl 2 , 90 MgCl 2 , and 5 HEPES, pH 7.2. Currents were measured using Axopatch 200B amplifier (Molecular Devices) at 30-mV pipette voltage in most experiments. The current and pressure traces were recorded simultaneously, and the analysis of activation midpoint pressures was done using PClamp-10 software. To ensure stability and constant midpoints of the excised patches, three linear ramp pulls were done before the addition of any peptide. For surviving patches, P1/P3 was added between the third and fourth pull and allowed to equilibrate for 15 min before pulls were resumed for a total of six measurements.